Context Engine

The Context Engine is Conduit's intelligence layer that builds persistent organizational knowledge from every interaction. While conversations expire, the patterns, expertise, and relationships discovered through them become permanent organizational memory.

Overview

Traditional enterprise systems treat each interaction as disposable. A question gets asked, an answer is given, and both disappear into chat logs no one will ever search. The Context Engine changes this by continuously extracting and structuring knowledge from every query, conversation, and collaboration that flows through Conduit.

The result is a living knowledge graph that knows:

• Who knows what across your organization
• How people collaborate and where bottlenecks exist
• What expertise exists and where gaps are forming
• How knowledge flows between teams and departments

Three Context Types

Conduit recognizes three distinct types of organizational context, each with different ownership and lifecycle characteristics.

1. Individual Context (Portable)

Individual context captures a person's demonstrated expertise, skills, and problem-solving patterns. This is built from their actual work -- queries they run, questions they answer, problems they solve.

What it includes:

• Expertise domains and proficiency levels
• Problem-solving approaches and patterns
• Query patterns and data source familiarity
• Response quality and depth metrics

Ownership: The individual. This context is fully portable -- it belongs to you and goes with you. You can export it at any time.

{
  "contextType": "individual",
  "userId": "eng-042",
  "expertise": [
    {
      "domain": "manufacturing",
      "level": "EXPERT",
      "score": 0.92,
      "evidenceCount": 847,
      "lastDemonstrated": "2026-02-09T14:30:00Z"
    },
    {
      "domain": "quality",
      "level": "HIGH",
      "score": 0.78,
      "evidenceCount": 234,
      "lastDemonstrated": "2026-02-08T09:15:00Z"
    }
  ],
  "portable": true,
  "exportFormats": ["json", "graphml", "csv"]
}

2. Relational Context (Company-Owned)

Relational context captures how specific people work together -- communication patterns, collaboration effectiveness, and knowledge transfer dynamics.

What it includes:

• Communication frequency and patterns between individuals
• Collaboration effectiveness scores
• Knowledge transfer direction and quality
• Response time patterns

Ownership: The company. Relational context is organizational IP. It naturally decays when one party leaves the organization, since the relationship no longer exists.

{
  "contextType": "relational",
  "participants": ["eng-042", "ops-017"],
  "metrics": {
    "collaborationFrequency": 12.5,
    "effectivenessScore": 0.87,
    "avgResponseTime": "4m 23s",
    "knowledgeTransferDirection": "bidirectional"
  },
  "decayOnSeparation": true
}

3. Organizational Context (Company-Owned)

Organizational context captures the structure and dynamics of teams, departments, and the organization as a whole.

What it includes:

• Org chart and reporting relationships
• Team structures and cross-functional connections
• Department expertise profiles
• Knowledge distribution maps

Ownership: The company. Organizational context is continuously updated as people join, leave, and move within the organization.

16 Expertise Domains

The Context Engine tracks expertise across 16 domains relevant to industrial operations:

| # | Domain | Description | | --- | ------------------ | ---------------------------------------------------------- | | 1 | Manufacturing | Production processes, throughput optimization, scheduling | | 2 | Engineering | Design, systems engineering, technical problem-solving | | 3 | Operations | Day-to-day plant operations, process control | | 4 | Quality | Quality assurance, SPC, defect analysis, compliance | | 5 | Safety | Safety protocols, incident analysis, regulatory compliance | | 6 | Maintenance | Preventive/predictive maintenance, reliability engineering | | 7 | Supply Chain | Procurement, logistics, inventory management | | 8 | Energy | Energy management, utilities, sustainability | | 9 | Environmental | Environmental compliance, emissions, waste management | | 10 | Automation | PLC/DCS programming, SCADA, control systems | | 11 | Data Analytics | Data analysis, reporting, visualization | | 12 | IT/OT | Network infrastructure, cybersecurity, system integration | | 13 | Planning | Production planning, capacity management, forecasting | | 14 | Regulatory | Regulatory compliance, auditing, documentation | | 15 | Training | Knowledge transfer, onboarding, skill development | | 16 | Leadership | Team management, decision-making, strategic planning |

Each domain is independently scored per individual, allowing the Context Engine to build detailed expertise profiles.

Composite Scoring

Expertise levels are computed using a composite scoring algorithm that weighs four factors:

| Factor | Weight | Description | | --------------- | ------ | -------------------------------------------------------------------- | | Frequency | 30% | How often the person demonstrates expertise in this domain | | Recency | 25% | How recently they demonstrated it (recent activity weighs more) | | Consistency | 20% | How reliable their contributions are over time | | Depth | 25% | How deep their expertise goes (answering complex vs. simple queries) |

The composite score is calculated as:

score = (frequency * 0.30) + (recency * 0.25) + (consistency * 0.20) + (depth * 0.25)

Scoring Example

Consider an engineer who frequently answers complex manufacturing queries:

Frequency:   0.88  (answers ~12 queries/week in this domain)
Recency:     0.95  (answered 3 queries today)
Consistency: 0.82  (reliable over past 6 months)
Depth:       0.91  (regularly handles complex, multi-step problems)

Composite = (0.88 * 0.30) + (0.95 * 0.25) + (0.82 * 0.20) + (0.91 * 0.25)
          = 0.264 + 0.2375 + 0.164 + 0.2275
          = 0.893 -> EXPERT level

Four Expertise Levels

Composite scores map to four expertise levels:

| Level | Score Range | Description | | ---------- | ----------- | -------------------------------------------- | | LOW | 0.00 - 0.25 | Basic familiarity, can handle simple queries | | MEDIUM | 0.26 - 0.50 | Working knowledge, handles routine questions | | HIGH | 0.51 - 0.75 | Strong expertise, handles complex scenarios | | EXPERT | 0.76 - 1.00 | Deep expertise, go-to person for this domain |

Levels are recalculated continuously as new evidence accumulates and temporal decay is applied.

Temporal Decay

To ensure context stays current and reflects the organization's actual state, the Context Engine applies temporal decay to all expertise scores.

How Decay Works

• Half-life: Configurable, default is 30 days
• Mechanism: Expertise scores decay exponentially based on time since last demonstration
• Effect: If someone stops demonstrating expertise in a domain, their score gradually decreases
• Reset: Any new demonstration of expertise resets the decay clock for that domain

decayed_score = score * (0.5 ^ (days_since_last / half_life))

Decay Example

An engineer with an EXPERT-level manufacturing score of 0.90 who takes a 60-day leave:

Day 0:   0.90  (EXPERT)
Day 30:  0.45  (MEDIUM)   -- one half-life
Day 60:  0.225 (LOW)      -- two half-lives
Day 61:  Re-demonstrates -> score begins rebuilding

This ensures that the Context Engine always reflects current, active expertise rather than historical credentials.

Context Export and Import

Individual context is portable and can be exported in multiple formats.

Supported Formats

| Format | Use Case | Description | | ----------- | -------------------- | ------------------------------------------- | | JSON | API integration | Structured data, full fidelity | | GraphML | Graph visualization | Nodes and edges for expertise relationships | | CSV | Spreadsheet analysis | Tabular format for reporting |

Export Example

# Export individual context as JSON
curl -X POST https://conduit.example.com/api/context/export \
  -H "Authorization: Bearer <token>" \
  -d '{"userId": "eng-042", "format": "json", "schema": "v1.1"}'

Response:

{
  "schema": "v1.1",
  "exportedAt": "2026-02-10T12:00:00Z",
  "userId": "eng-042",
  "contextType": "individual",
  "expertise": ["...omitted for brevity..."],
  "queryPatterns": ["...omitted for brevity..."],
  "auditLog": {
    "exportId": "exp-abc123",
    "requestedBy": "eng-042",
    "approvedBy": "system",
    "timestamp": "2026-02-10T12:00:00Z"
  }
}

Import

Context can be imported into a new Conduit instance when an individual joins a new organization:

curl -X POST https://conduit.example.com/api/context/import \
  -H "Authorization: Bearer <token>" \
  -F "file=@context-export.json" \
  -F "schema=v1.1"

All exports and imports are logged in the audit trail for compliance.

Find Experts

The Context Engine provides a real-time expert discovery API. Query it to find who knows what across your organization.

Basic Expert Query

# Find experts in manufacturing
curl https://conduit.example.com/api/context/experts \
  -H "Authorization: Bearer <token>" \
  -d '{"domain": "manufacturing", "minLevel": "HIGH"}'

Response:

{
  "domain": "manufacturing",
  "minLevel": "HIGH",
  "results": [
    {
      "userId": "eng-042",
      "name": "Sarah Chen",
      "level": "EXPERT",
      "score": 0.92,
      "department": "Production Engineering",
      "lastActive": "2026-02-09T14:30:00Z"
    },
    {
      "userId": "ops-089",
      "name": "Marcus Rivera",
      "level": "HIGH",
      "score": 0.71,
      "department": "Operations",
      "lastActive": "2026-02-10T08:45:00Z"
    }
  ],
  "cached": true,
  "cacheAge": "45s"
}

Multi-Domain Query

Find someone with expertise across multiple domains:

curl https://conduit.example.com/api/context/experts \
  -H "Authorization: Bearer <token>" \
  -d '{
    "domains": ["quality", "automation"],
    "minLevel": "MEDIUM",
    "matchAll": true
  }'

Caching

Expert queries are cached for performance:

• Cache TTL: 60 seconds (configurable)
• Cache invalidation: On new context evidence or score recalculation
• Real-time fallback: Cache miss triggers live graph query

How It All Works Together

Here is a conceptual flow of how the Context Engine builds knowledge:

User Activity                    Context Engine                     Knowledge Graph
-----------------                --------------                     ---------------

User asks NQE query         ->   Extract domain signal         ->  Update expertise score
                                 Identify query complexity          Adjust frequency metric

User answers a question     ->   Analyze answer quality        ->  Update depth metric
                                 Map domain coverage               Strengthen expert edges

Two users collaborate       ->   Track interaction pattern     ->  Create/update relational
                                 Measure effectiveness              context edge

User joins new team         ->   Update org structure          ->  Reorganize graph nodes
                                 Recalculate team profiles          Adjust organizational context

30 days pass (no activity)  ->   Apply temporal decay          ->  Reduce stale scores
                                 Recalculate levels                 Prune weak edges

The Context Engine operates continuously in the background. Every interaction that flows through Conduit -- whether it is an NQE query, a routed question, or a collaborative thread -- contributes evidence to the knowledge graph.

Next Steps

• AI-Mediated Collaboration -- See how the Context Engine powers intelligent question routing
• Privacy & Security Model -- Learn how context is abstracted and protected
• Natural Query Engine (NQE) -- Explore the data querying layer that feeds the Context Engine
• Architecture -- Understand the system design behind the Context Engine